The paper studies a particular type of waiting line system (a queue) where customers sometimes decide not to join, a behavior called "balking." The twist here is that the analysis is restricted to scenarios where exactly K customers do choose to join the queue within a specific time window. This kind of conditional analysis is useful in practice when you have after-the-fact knowledge about how many people used a service and want to understand what the experience was like during that period, including how many potential customers were turned away.
The main technical challenge is that tracking everything simultaneously, how many customers are in the system, how many have joined, and how many have balked, leads to an extremely large and complex mathematical model. The authors sidestep this by using a mathematical tool called a probability generating function, which compactly encodes the distribution of balking customers without requiring the full high-dimensional calculation. They also find a clever way to break the time window into two pieces and show that the probabilities in the second piece satisfy a simpler differential equation, making numerical computation much more manageable. For situations where arrival rates change over time in a step-by-step fashion, they develop a concrete procedure to compute key statistics.
Perhaps the most interesting part of the paper is the numerical examples, which reveal counterintuitive patterns. For instance, conditioning on a fixed number of joiners can distort your picture of what the system was actually doing, producing effects that run against common intuition. The authors carefully explain why these surprising behaviors arise, giving practitioners a clearer sense of how selection bias, introduced by knowing how many customers joined, can mislead conclusions about queue performance and customer behavior.